With the improvement of hardware computing power, the application of deep learning methods in the field of remote sensing is increasing. This paper summarizes the progress of deep learning methods in remote sensing image object detection in recent years. The main methods of deep learning methods to extract and use target feature information in various target detection tasks are summarized. Finally, the application trend of deep learning methods in the field of remote sensing image detection is prospected.
This paper extends the ground-level visual attributes to high resolution remote sensing imagery to demonstrate the useful-ness of visual attributes for remote sensing tasks such as image classification. Visual attributes have been introduced as the semantic properties that transcend the categories. We train predictors from the largest ground-level attributes datasets, SUN, for 102 visual attributes, which is well defined in SUN. We first form an attribute-based representation for the remote sensing imagery with the output of trained attribute predictors. We then evaluate the classification performances of the attribute-based representation against traditional features. Extensive experiments on the ground-level baseline dataset scene 15 and remote sensing dataset UCMLU shows that ground-level visual attributes outperform the traditional low-level features in the classification problem, and the combination of ground-level visual attribute and low-level features obtains best classification rate. Moreover, we demonstrate that attribute-based representation is much more semantically powerful than the low-level features.
Outliers and occlusions are important degradation in the real application of point matching. In this paper, a novel point matching algorithm based on the reference point pairs is proposed. In each iteration, it firstly eliminates the dubious matches to obtain the relatively accurate matching points (reference point pairs), and then calculates the shape contexts of the removed points with reference to them. After re-matching the removed points, the reference point pairs are combined to achieve better correspondences. Experiments on synthetic data validate the advantages of our method in comparison with some classical methods.
Detection of anomalous targets of various sizes in hyperspectral data has received a lot of attention in reconnaissance and surveillance applications. Many anomaly detectors have been proposed in literature. However, current methods are susceptible to anomalies in the processing window range and often make critical assumptions about the distribution of the background data. Motivated by the fact that anomaly pixels are often distinctive from their local background, in this letter, we proposed a novel hyperspectral anomaly detection framework for real-time remote sensing applications. The proposed framework consists of four major components, sparse feature learning, pyramid grid window selection, joint spatial-spectral collaborative coding and multi-level divergence fusion. It exploits the collaborative representation difference in the feature space to locate potential anomalies and is totally unsupervised without any prior assumptions. Experimental results on airborne recorded hyperspectral data demonstrate that the proposed methods adaptive to anomalies in a large range of sizes and is well suited for parallel processing.
Semantic classification of very high resolution (VHR) remote sensing images is of great importance for land use or land cover investigation. A large number of approaches exploiting different kinds of low level feature have been proposed in the literature. Engineers are often frustrated by their conclusions and a systematic assessment of various low level features for VHR remote sensing image classification is needed. In this work, we firstly perform an extensive evaluation of eight features including HOG, dense SIFT, SSIM, GIST, Geo color, LBP, Texton and Tiny images for classification of three public available datasets. Secondly, we propose to transfer ground level scene attributes to remote sensing images. Thirdly, we combine both low-level features and mid-level visual attributes to further improve the classification performance. Experimental results demonstrate that i) Dene SIFT and HOG features are more robust than other features for VHR scene image description. ii) Visual attribute competes with a combination of low level features. iii) Multiple feature combination achieves the best performance under different settings.
Image matching has always been a very important research areas in computer vision. The performance will directly affect the matching results. Among local descriptors, the Scale Invariant Feature Transform(SIFT) is a milestone in image matching, while HOG as an excellent descriptor is widely used in 2D object detection, but it seldom used as a descriptor for matching. In this article, we suppose to pool these algorithms and we use a simple modification of the Rotation- Invariant HOG(RI-HOG) to describe the feature domain detected by SIFT. The RI-HOG is Fourier analyzed in the polar/spherical coordinates. Later in our experiment, we test the performance of our method on a datasets. We are surprised to find that the method outperforms other descriptors in image matching in accuracy.
In this paper, we introduce and study a novel unsupervised domain adaptation (DA) algorithm, called latent subspace sparse representation based domain adaptation, based on the fact that source and target data that lie in different but related low-dimension subspaces. The key idea is that each point in a union of subspaces can be constructed by a combination of other points in the dataset. In this method, we propose to project the source and target data onto a common latent generalized subspace which is a union of subspaces of source and target domains and learn the sparse representation in the latent generalized subspace. By employing the minimum reconstruction error and maximum mean discrepancy (MMD) constraints, the structure of source and target domain are preserved and the discrepancy is reduced between the source and target domains and thus reflected in the sparse representation. We then utilize the sparse representation to build a weighted graph which reflect the relationship of points from the different domains (source-source, source- target, and target-target) to predict the labels of the target domain. We also proposed an efficient optimization method for the algorithm. Our method does not need to combine with any classifiers and therefore does not need train the test procedures. Various experiments show that the proposed method perform better than the competitive state of art subspace-based domain adaptation.
Automatic image registration is a vital yet challenging task, particularly for non-rigid deformation images which are more complicated and common in remote sensing images, such as distorted UAV (unmanned aerial vehicle) images or scanning imaging images caused by flutter. Traditional non-rigid image registration methods are based on the correctly matched corresponding landmarks, which usually needs artificial markers. It is a rather challenging task to locate the accurate position of the points and get accurate homonymy point sets. In this paper, we proposed an automatic non-rigid image registration algorithm which mainly consists of three steps: To begin with, we introduce an automatic feature point extraction method based on non-linear scale space and uniform distribution strategy to extract the points which are uniform distributed along the edge of the image. Next, we propose a hybrid point matching algorithm using DaLI (Deformation and Light Invariant) descriptor and local affine invariant geometric constraint based on triangulation which is constructed by K-nearest neighbor algorithm. Based on the accurate homonymy point sets, the two images are registrated by the model of TPS (Thin Plate Spline). Our method is demonstrated by three deliberately designed experiments. The first two experiments are designed to evaluate the distribution of point set and the correctly matching rate on synthetic data and real data respectively. The last experiment is designed on the non-rigid deformation remote sensing images and the three experimental results demonstrate the accuracy, robustness, and efficiency of the proposed algorithm compared with other traditional methods.
This paper proposes an algorithm of simulating spatially correlated polarimetric synthetic aperture radar (PolSAR) images based on the inverse transform method (ITM). Three flexible non-Gaussian models are employed as the underlying distributions of PolSAR images, including the KummerU, W and M models. Additionally, the spatial correlation of the texture component is considered, which is described by a parametric model called the anisotropic Gaussian function. In the algorithm, PolSAR images are simulated by multiplying two independent components, the speckle and texture, that are generated separately. There are two main contributions referring to two important aspects of the ITM. First, the inverse cumulative distribution functions of all the considered texture distributions are mathematically derived, including the Fisher, Beta, and inverse Beta models. Second, considering the high computational complexities the implicitly expressed correlation transfer functions of these texture distributions have, we develop an alternative fast scheme for their computation by using piecewise linear functions. The effectiveness of the proposed simulation algorithm is demonstrated with respect to both the probability density function and spatial correlation.
Displaced phase center antenna (DPCA) and along-track interferometry (ATI) are the two popular techniques used to determine synthetic aperture radar-ground moving target indication fields, and studies have shown that the combinations of these techniques can improve the target detection performance. However, a crucial problem is how to combine the two techniques, which requires a complete analysis and comparison of the individual techniques. Generally, it is well known that the performances of these techniques are closely related to clutter and noise. A detailed comparison of the detection performance of ATI and DPCA is presented, together with an assessment developed by theoretical analysis and simulations. The results show that the ATI is limited mainly by the clutter and noise, while DPCA is limited mainly by channel imbalance and noise. The ATI’s main drawback is its high false alarm rate, and DPCA is more sensitive to the channel imbalance. In most cases, DPCA is better than ATI, but for a high clutter-to-noise ratio, low signal-to-clutter power ratio, and channel imbalance, ATI has a better performance than DPCA. The real data experiments verify the theoretical findings. Meanwhile, the effects of target radial velocity, incidence angle, transmission bandwidth, and terrain type on the performance of the two detection approaches are also investigated.
In this paper, a novel local ways to implement hyperspectral anomaly detector is presented. Usually, the local detectors
are implemented in the spatial window of image scene, but the proposed approach is implemented on the windows of
spectral space. As a multivariate data, the hyperspectral image datasets can be considered as a low-dimensional manifold
embedded in the high-dimensional spectral space. In real environments, nonlinear spectral mixture occurs more
frequently. At these situations, whole dataset would be distributed in one or more nonlinear manifolds in high
dimensional space, such as a hyper-curve surface or nonlinear hyper-simplex. However, the majority of global and local
detectors in hyperspectral image are based on the linear projections. They are established on the assumption that the
geometric distribution of datasets is a linear manifold. It is incapable for them to deal with these nonlinear manifold data,
even for spatial local data. In this paper, a novel anomaly detection algorithm based on local linear manifold is put
forward to handle the nonlinear manifold problems. In the algorithm, the local neighborhood relationships are
established in spectral space, and then an anomaly detector based on linear projection is carried out in these local areas.
This situation is similar to using sliding windows in the spectral space. The results are compared with classic spatial
local algorithm by using real hyperspectral image and demonstrate the effectiveness in improving the weak anomalies
detection and decreasing the false alarms.
Inherent speckle noise greatly limits the performance of change detection in synthetic aperture radar (SAR) images. In order to reduce the influence of such speckle noise, a modified-log-ratio (MLR) operator for change detection of targets in forest concealment is proposed. By performing a logarithm of the linear transform ratio on the pixel amplitude using multilook processing to generate a difference image, the presented MLR operator is theoretically proven to be capable of improved restraining of the negative effect of speckle noise over the conventional log-ratio operator. A Gaussian distribution of the MLR statistics is assumed, which facilitates the adaptive detection of targets using a constant false alarm rate technique. Experimental results based on the public VHF-band CARABAS-II SAR image dataset validate the effectiveness of the proposed operator.
KEYWORDS: Statistical analysis, Hyperspectral imaging, Data modeling, Interference (communication), Error analysis, Principal component analysis, Signal to noise ratio, Data analysis, Vector spaces, Signal attenuation
Dimensionality Reduction (DR) for hyperspectral image data can be regarded as a problem of signal subspace estimation (SSE) in terms of the Linear Mixing Model (LMM). Most SSE methods for hyperspectral data are based on the analysis of second-order statistics (SOS) without considering preservation of anomalies. This paper addresses the problem of SSE for preserving both abundant and rare signal components in hyperspectral images. The multivariate sample skewness for testing normality is brought in our new algorithm as a discrimination index for rank determination of rare vectors subspace, combining with analysis of the maximum of data-residual ℓ2-norm denoted as ℓ2,∞-norm which is strongly influenced by the anomaly signal components. And the SOS based method, labeled as hyperspectral signal subspace identification by minimum error (HySime), is employed for identification of abundant vectors space. The results of experiments on real AVIRIS data prove that multivariate sample skewness statistics is suitable for measuring the distribution about hyperspectral data globally, and our algorithm can obtain the anomaly components from data that are discarded by HySime, which implies less information loss in the our method.
In this paper, a novel CFAR algorithm for detecting layover and shadow areas in Interferometric synthetic aperture radar (InSAR) images is proposed. Firstly, the probability density function (PDF) of the square root amplitude of InSAR image is estimated by the kernel density estimation. Then, a CFAR algorithm combining with the morphological method for detecting both layover and shadow is presented. Finally, the proposed algorithm is evaluated on a real InSAR image obtained by TerraSAR-X system. The experimental results have validated the effectiveness of the proposed method.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.